Parameter Estimation in The Huff Model: Pij SJ Dij / SJ Dij

Parameter Estimation in the Huff Model
By David L. Huff
Editorʼs note: Dr. David L. Huff pioneered several spatial analysis techniques, most notably the Huff Model, which was introduced in 1963 in
an article in Land Economics entitled “A Probabilistic Analysis of Shopping Center Trade Areas.” The model has been used for tasks such as
predicting consumer spatial behavior, delineating trade areas, locating retail and service facilities, analyzing market performance, simulating
different market scenarios, and forecasting sales.
The Huff Model has endured the test of selling area to the distance from a consumerʼs out an accurate statistical assessment, analysts
time—more than 40 years. Its widespread residence to the store. Each of these variables are using the model to make predictions based
use by business and government analysts, as was weighted by an exponent (i.e., parameter)
well as academicians, throughout the world is that was estimated empirically by surveying
remarkable. With the development of GIS, the the shopping preferences of individuals in the
model has received even more attention. study area.
The modelʼs longevity can be attributed to It was posited that the size of the store was
its comprehensibility, relative ease of use, and more important for some products than oth-
its applicability to a wide range of problems. ers. Consequently, the value of the exponent
The model is conceptually appealing. Its logi- could be expected to be larger for these prod-
cal underpinning makes sense, and the output ucts. Conversely, the exponent for distance
can be communicated easily and understand- was assumed to be negative. Convenience
ably. The model is relatively easy to make op- products could be expected to have a larger
erational because the necessary computations exponent while specialty goods would have a
are straightforward once the values of the much smaller exponent. The probability that a
variables and the parameters have been speci- consumer located at i selecting store j can be
fied. Finally, its applicability to a wide range estimated using the formula in Figure 2. In this
of problems and its ability to predict outcomes
that would be difficult, if at all possible, with-
out the model have made its use widespread. Pij =Sj α Dij ß / ∑Sj α Dij ß
However, despite the general applicability Figure 2: Consumer store selection
of the model, it has not always been employed
correctly and its full potential has not been re- formula, Pijij is the probability of a consumer
alized. After a quick review of the model, the located at i choosing store j.. S Sj is the square
remainder of this article will address these two footage of store j; Dij ij is the distance from i
issues. to j; and a and b are parameters that were es-
timated based on the actual survey data. This
Review of the Model calculation was done using an approximation
The model is based on the premise of the solution since the application of conventional
probability that a person confronted with a statistical procedures was considered impos-
set of alternatives will select a particular item sible.
is directly proportional to the perceived util- Once the parameters were estimated, not
ity of each alternative. Choice behavior can only could the probabilities of patronage be
be viewed as probabilistic. As a result, it is estimated, but also the expected purchases
unlikely that any given alternative will be se- from subareas within the study area as shown
lected exclusively unless no other alternatives in Figure 3. In this scenario, Eij is the ex-
exist. This proposition can be expressed sym- pected purchases from area i to store j and Bik
bolically as shown in Figure 1. The probability is the amount budgeted by consumers in i for
that an individual (i) will select the alternative product k.
n
=Uj / ∑ Uj
Pij =Uj U Eij =Pij Bik
Figure 3: Patronage and purchases
j =1
Figure 1: Choice behavior Lack of Statistical Verification
Analysts using the Huff Model typically in-
(j) given the utility of j is relative to the sum of corporate some measure of accessibility (e.g.,
the utilities of all other choices (n) considered road distance, travel time, cost) as well as a
by individual (i). In order for the model to be variable that reflects the attraction of a given
applied, the utility of each alternative must be destination. The weights (i.e., parameters asso-
defined empirically. ciated with these variables) are often assigned
The first geographic application of the arbitrarily. They are rarely estimated statisti-
model was an attempt to predict consumer cally. As a result, the statistical significance
patronage patterns for different classes of of these variables is unknown. Obviously the
products. The utility of a store was defined as lack of statistically validated variables and pa-
the ratio of the square footage of the storeʼs rameters can produce erroneous results. With-
34 ArcUser October–December 2003 www.esri.com

Developer’s Corner
on unverifiable inputs. As a consequence, the perceived by some users to be much more dif- tive Marketing Effectiveness, 1988), have been
results are subject to error. ficult to calibrate because the model must be interested in this problem. This has resulted in
linearized with respect to its parameters before major breakthroughs that now make it possible
Why Calibration Is Omitted standard statistical estimation methods can be to use standard techniques such as ordinary
There are three reasons why more analysts applied. A number of researchers, including least squares for calibrating the model.
do not calibrate the model statistically. First, Lee G. Cooper and Masao Nakanishi (see The necessity (and the difficulty) of in-
the nonlinear properties of the model are Market-Share Analysis: Evaluating Competi- Continued on page 36
The Huff Model has been widely used for predicting

consumer spatial behavior, delineating trade areas,
locating retail and service facilities, analyzing
market performance, simulating different market
scenarios, and forecasting sales.
www.esri.com ArcUser October–December 2003 35

Parameter Estimation in the Huff Model
Continued from page 35
corporating origin-based data as opposed to number of executives understand and appre- ing adopted by more businesses is a logical
destination-based data supplies another reason ciate the value of thematic maps and address and defensible cost-benefit analysis of its ben-
why the model is often not calibrated statisti- matching and the importance of trade area efits and an appreciation that the bottom line is
cally. Actual choice behavior (the variable to analysis, customer profiling, and target mar- what matters in business.
be predicted) must be obtained empirically keting. This information is ideally suited for
from residents of geographic subareas located display on maps and can be related to without
within some larger study area, and choices difficulty.
must be obtained for all alternatives consid- The benefits of modeling are not always
ered by these residents. Unfortunately, most recognized by business decision makers—it
companies collect data only on their facilities is different and much more difficult. Deci-
(i.e., destination data). As a result, patronage sion makers, in general, lack the proficiency
data for competitors is usually unknown. The to comprehend the technical aspects of model
task of obtaining actual shopping preferences development and use. In addition, most soft-
for different products can be time-consuming, ware distributors do not realize that companies
as well as expensive, because it requires sur- do not want to spend money for research and
veying at the household level. development. What is required to get model-
The lack of GIS software packages that are
equipped for this type of analysis also inhibits
the correct application of the model. To per-
form this type of analysis, the software must
• Be able to execute the necessary operations
of the model. Huff Takes Advisory Role at ESRI
• Include a statistical package that can gen-
erate the statistics necessary for assessing the Dr. David L. Huff, an internationally recognized
significance of variables used to predict choice researcher in the spatial aspects of marketing and cre-
behavior and indicating how well the model ator of the Huff Model, has joined the ranks of ESRI
predicted actual choice frequencies.
• Possess mapping capability that enables technical contributors. As part of his exclusive agree-
analysts to examine errors of prediction as ment with ESRI, he will serve as an advisor and use his
well as other geographic patterns that might expertise to develop advanced predictive models and
be suggested by the data.
• Be designed for use by non-GIS specialists review existing analysis capability in ESRI’s business
as a decision making tool. analysis product and service offerings.
Designed to predict consumer spatial behavior, the
Enhanced Capabilities of the Model
Huff Model is used extensively by market analysts and
As mentioned previously, many variables in
the Huff Model can be included and the as- planners and has served as the seminal work for oth-
sociated parameters determined statistically ers doing research in modeling choice behavior.
so that the impact of a variety of variables As a professor at the University of Texas, Austin, Huff has held a variety of
can now be assessed that would not have
been possible previously. In general, predic- administrative and teaching positions in both geography and marketing and has
tive variables can be classified as being either consulted for dozens of agencies including the United States Department of Trans-
controllable or noncontrollable. portation; United States Bureau of the Census; the National Institutes of Health; Na-
Controllable variables, such as advertising
tional Endowment for the Arts; Resources for the Future, a nonprofit and nonparti-
and promotion, pricing, and store format, can
be controlled or influenced by the company. san think tank; numerous state and regional offices; and hundreds of businesses.
Noncontrollable variables, such as acces- Other research areas that Huff has contributed to include the application of mul-
sibility, population distribution, income, and
tivariate graphic displays to market analysis, the formulation of objective measures
competition, are typically beyond the firmʼs
control. Traditionally, some of these variables for delineating market areas, the development of computerized systems to monitor
are examined visually on thematic maps or pin economic activities geographically, and the derivation of planning regions for the
maps and subjective conclusions advanced geographic delivery of health care and economic services.
as to the probable impact of these variables.
However, being able to examine these vari- In 1988, Huff was awarded the James R. Anderson Medal in Applied Geography,
ables statistically adds immeasurably to the the highest honor awarded by the Association of American Geographers for distin-
accuracy of such conclusions. guished research in applied geography. He also received the Distinguished Mentor
Award from the National Council for Geographic Education in 1998 for his many
Increasing Business Usage
Most executives have not been shown how years helping students pursuing master’s and doctoral degrees.
their revenues could increase by investing in
software, modeling, and data. An increasing
36 ArcUser October–December 2003 www.esri.com

Parameter Estimation in The Huff Model: Pij SJ Dij / SJ Dij

Uploaded by

Copyright:

Available Formats

Parameter Estimation in The Huff Model: Pij SJ Dij / SJ Dij

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Parameter Estimation in The Huff Model: Pij SJ Dij / SJ Dij

Uploaded by

Copyright:

Available Formats

Parameter Estimation in the Huff Model

34 ArcUser October–December 2003 www.esri.com

The Huff Model has been widely used for predicting

www.esri.com ArcUser October–December 2003 35

You might also like